January 11, 2019

}

Quantity ordered and Amount spent have high variability

Overall, quantity ordered and amount have high variability. Also, quantity and amount aren't correlated with each other

Which lead to high variability in Price Per Unit


As expected, Price Per Unit also has high variability across months

80% of amount was spent on top 3 item codes

Summary of spent and quantity by item code
Item_Code Total_Spent Total_Quantity Spent_Percent Quan_Percent
BOX $2,416,795 167,239 31% 8%
BOT $1,892,747 99,611 24% 4%
RIS $1,257,950 196,213 16% 9%


Around 80% of spend is on top 3 items and these top 3 items correspond to approximately 20% of total quantity ordered

Even at item code level, there is high variability in Price Per Unit


Price per unit has high variation for BOT and BOX. Variation is considerably less for RIS

Splitting further to understand variation at Species level

This table shows number of species where coefficeint of variation at species level is greater or lesser than overall coefficient of variation (i.e. at Item code level)

Summary of coefficient of variation
Item_Code Num_Species_Greater Num_Species_Lesser Greater_Percent
BOT 9 8 53%
BOX 0 14 0%
RIS 7 10 41%


  • For all species in BOX, variation in price is less than overall price. Hence variation here could be because of order strategy

  • For BOT around 53% species have variation greater than overall and for RIS around 41% have variation greater than overall

Around 50% of species from RTR have higher variability

Variation across vendors
Vendor_Name High_CV_Species Low_CV_Species %High_CV_Species High_CV_PPQ Low_CV_PPQ
CFFCO USA INC 1 4 20% $20 $14
RTR Wood Products, Inc 13 14 48% $16 $9
WISSMAN BROTHERS, INC 1 0 100% $78 $0
Young Manufacturing Company, Inc. 2 8 20% $36 $16


  • Around 50% of species ordered from RTR have higher variability and same number is 20% from CFFCO

  • Price Per Unit is higher in high variability Species indicating indicating higher the unit price, higher the variation

Built various Machine Learning Models to predict price

Below are important metrics considered

  • Quantity
  • Item Code
  • Species
  • Vendor
  • Time taken to close order

Below are various Machine Learning Models considered

  • Linear Regression
  • Decison Trees
  • Random Forest
  • Gradient Boosting Machines

Able to predict expected price with average deviation of 15%

  • Built around 4k models to choose best performing model
  • Built an automated dashboard which gives expected price as output
  • Dashboard demo

Next Steps

  • Settingup Dashboard for production
  • Testing Model
  • Model in to prodcution